SmultiSNE: S-Multi-SNE: Semi-supervised multi-SNE

In theorod93/multiSNE: Multi-SNE: Multi-view dimensionality reduction

S-Multi-SNE: Semi-supervised multi-SNE

Description

This function performs S-multi-SNE, a visualization and classification algorithm for multi-view data. One of the data-views in the input list contains the labeling information with dummy variables representing the different clusters. Unlabeled samples are represented with NA in all variables of the labeling data-view.

Usage

SmultiSNE(
  X,
  initial_config = NULL,
  k = 2,
  initial_dims = 30,
  perplexity = 30,
  max_iter = 1000,
  min_cost = 0,
  epoch_callback = NULL,
  whitening = FALSE,
  epoch = 100,
  weights = NULL,
  weightUpdating = TRUE,
  lambdaParameter = 1
)

Arguments

X	A list with each element representing a data-view. All data-views must have the same number of rows.
initial_config	Initialization matrix, specifying the initial embedding for each data-view. Should be a list of the same length as X. Default is NULL.
k	Number of dimensions in the lhydatent embeddings. Default is 2.
initial_dims	The number of dimensions of the whitened data-views, if whitening True. Default is 30.
perplexity	Value of perplexity parameter. Default is 30.
max_iter	Maximum number of iterations to run. Default is 1000.
min_cost	Minimum cost to halt iteration Default is 0.
epoch_callback	A callback function used after each epoch (an epoch here means a set number of iterations). Default is NULL.
whitening	If True, whitening process to reduce the dimensions of the input data-views will be applied prior to multi-SNE. Default is FALSE.
epoch	The number of iterations in between update messages. Default is 100.
weights	Initialization of the weights. A vector of the same length as X. Default is NULL.
weightUpdating	Boolean. If True, weights will be updated at each iteration. Default is FALSE.
lambdaParameter	Parameter to indicate the contribution of weight update. Default is 1.

Value

Y : The latent embeddings.

Weights : A matrix containing the weights used for each iteration. Rows represent the iteration and columns the data-view.

Errors : A matrix containing the errors used for each iteration. Rows represent the iteration and columns the data-view.

Examples

# Get sample data
X <- vector("list")
X$first_dataView <- rbind(matrix(rnorm(10000),nrow=500,ncol=20), matrix(rnorm(5000, mean=1,sd=2),nrow=250,ncol=20))
X$second_dataView <- rbind(matrix(rpois(20000, lambda = 1),nrow=500,40), matrix(rpois(10000, lambda=3),250,40))
X$labeling <- rbind(cbind(rep(1,500), rep(0,500)), cbind(rep(0,250), rep(1,250)))
nan_index <- sample(seq(1,750,1),250)
X$labeling[nan_index,] <- c(NA,NA)
# Run S-multi-SNE
Y <- SmultiSNE(X, max_iter = 200)
true_labels <- c(rep(2,500), rep(3,250))
missing_labels <- true_labels
missing_labels[nan_index] <- 1
par(mfrow=c(1,2))
plot(Y$Y, col = true_labels, pch=19)
plot(Y$Y, col = missing_labels, pch=19)

theorod93/multiSNE documentation built on April 22, 2024, 11:16 a.m.